AI prompting for marketers—agency leaders share lessons for generating creative ideas

For the past two years, marketers have been learning how to prompt AI systems by throwing ideas at the wall and seeing what sticks. This approach has yielded a few concrete strategies, such as step-by-step formulas to condense prompt information and galleries that house a number of text templates. But sometimes the black box of AI calls for the use of surprising measures to achieve desired results. Case in point: When Jesse Suchmann is faced with a raft of complicated data, he’ll sometimes ask a chatbot to break it down in the style of a drunk friend at a bar. “That always gets me straight to the point,” said Suchmann, an executive VP and executive creative director at Edelman. And if the bot happens to generate some harsh language, Suchmann said, then summaries will be more memorable. Check out the latest Ad Age events and award programs This is one of a handful of hacks that marketers are using to level up their AI prompting game for chatbots including ChatGPT, Gemini and Claude. Prompts are the way that users query AI systems, and are typically constructed as pieces of text. But as with any form of writing, prompting is an art, and should therefore be intentionally guided to produce the best results. Below, Ad Age has compiled some of these tactics from agency personnel who tinker with AI daily. Whether you’re getting started with a prompt or putting on the final touches, here is what the experts recommended. Don’t be stingy with detail “I think the main thing to consider is that gen AI needs and deserves the same input as a human copywriter to write good copy,” said Bas Korsten, VML’s global chief creative officer of innovation and chief creative officer, Europe, Middle East and Africa, via email. “The more context (e.g. client brief, creative brief, key insights, brand guidelines, tone of voice, etc.), the more detail, the more guidance, the better.” Jessica Berger, senior VP of innovation at Publicis Media, also provides extensive details from the start. “Typically, I provide a short concept description, followed by copy prompts tailored to different audiences or sentiments,” Berger said. “If we have visuals like images or videos, the prompts align with those storyboards to maintain consistency.” Berger uses a variety of prompting structures to prime the AI for different tasks, including telling the AI to “act as [blank]” in order to give it a role to inhabit. Berger also gives the chatbot context, such as telling it to “write for” a specific scenario, as well as guardrails, such as telling it to incorporate a specific tagline or avoid jargon. These direct instructions help Berger avoid broad prompts, which are an easy way to waste a lot of one’s time. “A far too broad and unfocused prompt looks something like, ‘Tell me everything about creating an ad for a new soft drink,’” Berger said. “Compare this to: ‘Develop a tagline and short ad copy for a new soft drink that’s designed for gamers. Focus on its high caffeine content and sleek design. Include two versions: one for a competitive esports audience, and one for casual gamers.’” Some context is unhelpful “Overloading the AI with too many questions or complex context often leads to unclear or convoluted results,” Berger said. “It's better to focus on one task at a time to keep outputs clean and streamlined.” When chatbots are faced with confusing context, they may hallucinate, which is why Berger also reinforces factual accuracy in her prompts by specifying: “Only reference real people or well-documented events,” Berger said. Also read: AI in ads divides consumers AI is not Don Draper No matter how hard a marketer may try, the current AI chatbots aren’t going to yield great advertising content on their own. Some platforms, including ChatGPT, have yet to nail down the nuances between basic marketing concepts, said Emma LeMay, a copywriter for Pereira O’Dell. For instance, chatbots might not know the distinction between a “headline” and a “tagline,” LeMay said. Asking a bot to write ads will generate only the most predictable ideas, said Peter Steiner, senior director of creative technology at Code and Theory. This is because AI is essentially predictive, using past data to surface expected content. It is more useful to think of models, such as ChatGPT, as interpretive AI, not generative AI, Steiner said. “The technology works by reorganizing or recontextualizing information,” Steiner said. “So it doesn’t really generate anything. At best, it reveals new connections you didn’t see.” Also read: How AI would write Super Bowl commercials When in doubt, start again “Start over constantly,” Steiner said. “If you’ve been refining the prompt for a while, and you don’t know what to do next, hit the reset button. Start a new chat with your refined prompt. … From a purely technical point of view, you’ve got to clear the ‘memory’ so the model doesn’t get confused or distracted by any old ideas that you’ve abandoned. Start fast and fail faster.” The idea of failing fast may sound unpleasant to some, but there are ways to depersonalize the experience and get it moving even more quickly. For example, Steiner often asks the model to write the prompt for him. “That will quickly get you past the blank page, and often it gives you a good framework to start with,” Steiner said. And while if doing so may initially yield poor responses, even poor responses could be useful. “Start a new conversation and give the AI the prompt plus the bad response,” Steiner said. “Tell it what you don’t like and ask it to help you fix the prompt.” “I also love to ask [the bot], ‘What am I missing’ or ‘What am I not thinking of,’” Steiner added. “The AI might surprise you.” See Ad Age’s 2024 Breakout Brand Leaders and Work here Try on different audiences “Always look to fine-tune the tone, energy, and call-to-action based on the audience's mindset, ensuring the final product is tailored and resonates,” Publicis’ Berger said. An example Berger gave: “Rewrite the following copy about a tech gadget in a style that evokes curiosity and excitement. Then create a version that appeals to tech skeptics, focusing on trust and reliability.” Pereira O’Dell’s LeMay takes a similar approach, feeding the AI pre-existing work and requesting adjustments to tone, such as in the voice of an eight-year-old, LeMay said. VML’s Korsten prefers to create distinct AI personas with different viewpoints, which can provide feedback that varies from persona to persona. Providers including Google have begun to make persona creation easier through single-purpose AI assistants. Trim the fat “Character count is an ongoing battle for copywriters,” LeMay said. Fortunately, LeMay has found ChatGPT to be effective taking headlines or copy she has written and fitting them to new character lengths. For Edelman’s Suchmann, only when the copy feels ready to go does he turn to the bot’s trimming abilities. “My favorite prompt is asking it to cut the length by 20% after I think it’s perfect,” Suchmann said. “That usually trims just the right amount of fat, then I’ll add back anything fun that I miss.”

AI prompting for marketers—agency leaders share lessons for generating creative ideas

Featured Stories